Salesforce already employs agent managers. Not AI agents that manage things, but humans whose full-time job is managing fleets of AI agents. Zach Stauber, a support agent manager at Salesforce, described his daily routine in a February 2026 HBR article: “Data, Data, Data. I start and end my day in dashboards, scorecards, and agent observability.” His job looks less like a software engineer’s and more like a floor manager’s, except the floor is digital and the team is made of LLMs.
This is not a theoretical role. Salesforce’s Agentforce platform autonomously resolves nearly 74% of inbound support cases. Their SDRs went from 150 meetings in 30 days to 350+ meetings per week after deploying AI agents, generating $60 million in annualized pipeline and acquiring 300+ new clients in four months. Someone has to oversee that. Someone has to notice when resolution quality dips, when a prompt needs refinement, or when an agent should escalate instead of hallucinate an answer. That someone is an agent manager.
Why This Role Exists Now
The World Economic Forum projects that 39% of workers’ core skills will change by 2030, with AI fluency topping the list. Gartner predicts 40% of enterprise software applications will include task-specific AI agents by 2026, up from under 5% in 2024. At that scale, the question stops being “should we deploy agents?” and becomes “who is responsible for them once they’re running?”
Most companies answer this question badly. They treat agents like software: deploy, monitor logs occasionally, maybe assign a Slack channel. That approach breaks because agents are not deterministic. They make decisions. They interact with customers. Their behavior shifts when models update. Research compiled by Second Talent found that over 50% of enterprise AI tools operate as unsanctioned “shadow agents” with no clear owner.
The agent manager role fills this gap. Just as product managers emerged during the software era to bridge engineering and business, agent managers bridge AI capabilities and business outcomes. HBR’s authors, Harvard Business School professor Suraj Srinivasan and Salesforce COO Vivienne Wei, predict that “agent manager” will become a standard title within 12-18 months in AI-first enterprises.
The Product Manager Parallel
Product managers do not write code. They define what should get built, prioritize features, and translate customer needs into engineering requirements. Agent managers follow a similar pattern: they do not build models, but they define what agents should do, set performance targets, refine prompts and workflows, and translate business goals into agent behavior.
The difference is speed. A product manager ships quarterly. An agent manager operates in weekly test-deploy-learn cycles because prompts can be updated, evaluated, and pushed in hours, not months.
What Agent Managers Actually Do
The HBR article outlines a set of daily responsibilities that look nothing like a traditional engineering or management role. Here is what fills an agent manager’s day.
Performance Monitoring and Diagnostics
Agent managers track metrics across quality, speed, escalation rates, and customer sentiment. When Salesforce’s Agentforce handles 74% of support cases autonomously, the remaining 26% are the ones that matter most. An agent manager digs into why those cases needed human escalation: was the agent’s knowledge base incomplete? Did it misinterpret intent? Was the prompt too narrow?
This is diagnostic work. It looks like QA, but it requires understanding both the business domain and how LLMs process context. You cannot debug an agent’s behavior if you do not understand the customer journey it is supposed to serve.
Prompt and Workflow Refinement
The biggest lever an agent manager pulls is prompt engineering, but at an operational level, not a research one. When a support agent starts giving weaker answers on a specific product line, the agent manager identifies the gap, adjusts the grounding data or prompt structure, tests the change against historical cases, and deploys it. This cycle happens weekly.
Vanessa Tabbert, VP of Agentic Transformation at Salesforce, emphasized in the HBR article that the best agent managers treat prompts as living documents that evolve alongside the business.
Human-Machine Handoff Design
Every agent fleet needs escalation paths. The agent manager defines when an agent should stop trying and route to a human, what context it should pass along, and how the handoff experience feels to the customer. This is where domain expertise matters. An agent manager for a financial services firm needs to understand which customer scenarios carry regulatory risk. An agent manager in healthcare needs to know which symptom combinations require immediate human review.
ROI Analysis and Executive Reporting
Agent managers own the business case. They quantify how much the agent fleet saves, how it affects customer satisfaction scores, and where additional investment would yield the highest return. At Salesforce, the numbers are concrete: $60 million in pipeline generated, 300+ new clients, SDR productivity multiplied. An agent manager does not just report these numbers; they attribute them to specific agent improvements and make the case for scaling or restructuring the fleet.
Six Skills Every Agent Manager Needs
The HBR article identifies six core capabilities. They map to a unique blend of technical literacy, domain expertise, and operational agility.
1. AI Operational Literacy. You do not need to train models. You need to understand how agents process context, when they hallucinate, how retrieval-augmented generation works, and what agent observability tools measure. Think of it as knowing enough about the engine to diagnose problems, without needing to rebuild it.
2. Functional Depth. Generic AI knowledge is not enough. An agent manager for a sales team needs deep sales process knowledge. An agent manager for customer support needs to understand support workflows, escalation policies, and service-level agreements. The domain is where the edge is.
3. Systems Thinking. Most agent deployments involve multiple agents working together. A support agent hands off to a billing agent, which triggers a notification agent. The agent manager needs to see the full chain and identify where breakdowns happen. Multi-agent orchestration is already complex; managing it operationally is harder still.
4. Change Resilience. Models update. APIs change. Customer expectations shift. Agent managers operate in weekly cycles. The HBR authors describe a “test-deploy-learn” cadence that requires comfort with constant iteration. If you need a stable, predictable workflow, this is not the role.
5. Prompt Craftsmanship. This is not about writing clever prompts for ChatGPT. It is about designing behavioral language and decision logic that govern how agents operate at scale. One word in an escalation prompt can mean the difference between a 60% and 80% autonomous resolution rate.
6. Hybrid Workflow Design. Building processes where humans and agents collaborate without friction. This means designing escalation triggers, fallback procedures, and feedback loops where human input improves agent performance over time.
Who Gets Hired for This Role
Here is the most counterintuitive finding from the HBR research: effective agent managers rarely come from pure AI backgrounds. Stauber’s background is in audio production and conversational design. The skills that matter most are accountability for service quality and customer outcomes, not machine learning expertise.
HR Executive reports that organizations should treat AI agent deployment like hiring: clear job descriptions, designated supervisors, defined performance expectations, and built-in review mechanisms. The people who excel at those processes, operations leads, service managers, customer experience directors, are the ones transitioning into agent management.
Workers with AI skills command wage premiums up to 56% higher than peers without them. For agent managers specifically, the premium is likely higher because the role combines AI fluency with domain expertise and people-management instincts.
Companies already hiring for the role include Salesforce (support agent manager, VP of Agentic Transformation), JPMorgan Chase, and Walmart. Job postings are also appearing under titles like “AI Agent Operations Lead,” “Agentic Workflow Manager,” and “Head of Agent Experience.”
How to Build an Agent Management Function
If you are setting up agent management inside your organization, here is a practical sequence.
Start With One Fleet, One Manager
Do not hire a “VP of Agent Management” before you have a single agent fleet running. Pick the business function where AI agents have the most impact (customer support is the typical starting point) and assign one person who owns agent performance end-to-end. This person should already have domain expertise in that function.
Define the Agent’s Job Description
HR Executive’s guidance applies directly: every AI agent should have a written scope of work, just like a new hire. What can it do? What can it not do? When should it escalate? What metrics define success? The agent manager owns this document and revises it as the agent improves.
Build the Observability Stack
Agent managers need dashboards. At minimum: resolution rate, escalation rate, customer satisfaction per interaction, average handling time, and error categorization. Salesforce’s Stauber relies on scorecards and agent observability tools built into Agentforce. If you are building on open-source frameworks, tools like LangSmith, Arize Phoenix, or Helicone provide similar capabilities.
Establish the Feedback Loop
The test-deploy-learn cycle only works if agent improvements are measured against historical performance. Track every prompt change, A/B test where possible, and document what worked and what did not. This institutional knowledge compounds over time and becomes the agent management team’s most valuable asset.
Frequently Asked Questions
What is an AI agent manager?
An AI agent manager is a human leader responsible for overseeing fleets of AI agents. They monitor agent performance, refine prompts and workflows, design human-machine handoffs, and report ROI to executives. The role parallels how product managers bridge engineering and business, except agent managers bridge AI capabilities and business outcomes.
What skills does an agent manager need?
According to HBR research, agent managers need six core skills: AI operational literacy, deep functional domain expertise, systems thinking for multi-agent orchestration, change resilience for weekly iteration cycles, prompt craftsmanship at an operational level, and hybrid workflow design for human-agent collaboration.
Do agent managers need a technical AI background?
Not necessarily. HBR found that effective agent managers often come from operations, customer experience, or service management backgrounds rather than pure AI or machine learning roles. The most important qualifications are accountability for service quality and domain expertise, combined with enough AI literacy to understand how agents process context and where they fail.
Which companies are hiring agent managers?
Salesforce already employs agent managers and VPs of Agentic Transformation. JPMorgan Chase and Walmart are also building agent management functions. Job postings appear under titles like AI Agent Operations Lead, Agentic Workflow Manager, and Head of Agent Experience. HBR predicts “agent manager” will become a standard title within 12-18 months at AI-first enterprises.
How is an agent manager different from a product manager?
Product managers define what gets built and ship quarterly. Agent managers define how AI agents behave and operate in weekly test-deploy-learn cycles. Both translate business goals into system behavior, but agent managers work at operational speed because prompt and workflow changes can be deployed in hours. Agent managers also own performance monitoring, escalation design, and human-agent handoff processes.